During the past several years, many CAD schemes in mammography were developed and tested with varying degrees of success. For the most part, these systems have been evaluated by measuring performance on limited sets of data, and in some cases, on the training set itself, with little effort to assess how performance would be effected if the algorithms were applied to a large data base or preferably the general image ensemble. Since the diagnostic performance of some CAD systems is approaching the point where clinical utility can be demonstrated, it is becoming increasingly important to explore ways not only to optimize these schemes, but also to assess their robustness when used in the clinical environment. In this project, we propose to conduct a variety of studies to improve and test the performance and robustness of two of our own CAD schemes. We will also assess a number of issues associated with system's performance when multiple images (e.g., views) and/or multiple independent schemes are available and are used to optimize the results of a diagnostic examination.